110kV Cable Joint Temperature Computation Based on Radial Basis Function Neural Networks

It is significant for engineering practice to monitor the hot spot temperature of cable joint which is a weak link in the transmission line. For this purpose, a computation model of cable joint temperature was established by radial basis function neural networks in this paper, in which square of current, surface temperature of prefabricated rubber and ambient temperature at present and before several hours called delay time were taken as inputs and real-time cable joint temperatures were outputs. The effects of model parameters were analyzed through finite element simulation of temperature field, and the focus was drawn to the determination of delay time which was approximately equal to three times as long as the time lag of prefabricated rubber temperature. In order to verify this algorithm, 110kV cable joint temperature rise test was carried out in the laboratory with multi-amplitude step current. The computation temperature based on radial basis function neural networks was in good agreement with the test result showing a high precision of this model, and the optimal delay time was pretty close to triple the time lag of prefabricated rubber temperature consistent with theoretic analysis by simulation. This research contributes to improving the computation accuracy of cable joint temperature and has great significance for assessing the insulation state of cable joint.